a review on condition monitoring and diagnostic techniques of rotating electrical machines

____________________________________________________________________________________________

*Corresponding author: Email: [email protected];

Physical Science International Journal4(3): 310-338, 2014

SCIENCEDOMAIN internationalwww.sciencedomain.org

A Review on Condition Monitoring andDiagnostic Techniques of Rotating Electrical

Machines

S. A. Mortazavizadeh1 and S. M. G. Mousavi1*

1Iran University of Science and Technology (IUST), Tehran, Iran.

Authors’ contributions

Author SAM wrote the first draft of the paper and author SMGM read and approved the finalmanuscript.

Received 11th May 2013Accepted 20 June 2013

Published 19th November 2013

ABSTRACT

Electrical machines are critical components in industrial processes. A motor failure mayyield an unexpected interruption at the industrial plant, with consequences in costs, productquality, and safety. To determine the conditions of each part of motor, various testing andmonitoring methods have been developed. In this paper, a review on effective faultindicators and condition monitoring methods of rotating electrical machines has beenaccomplished. Fault detection methods divided to four groups: electrical, mechanical,chemical and thermal indicators. Some fault detection methods based on electricalsymptoms like stator current, voltage, their combination or spectrum discussed in electricalgroup. In second branch, mechanical symptoms like torque, vibration and so on used forcondition monitoring. Third group, chemical indicators, assigned to some chemicalparameters of materials like oil characteristic or wear and debris in oil analysis. In last group,thermal symptoms in rotating electrical machines will be spoken. Between all methods,some of them are more known like vibration and some of them are recently added like motorcurrent signature analysis (MCSA). Nowadays, combined methods and methods usedartificial intelligence (AI) in condition monitoring are more popular. In every group, the faultdetection method and the faults that can be detected have been mentioned. Mathematicalequations of some new signal processing method have been discussed in literaturepresented in appendix.

Review Article

Physical Science International Journal, 4(3): 310-338, 2014

311

Keywords: Condition monitoring; electrical motor; fault diagnosis; review.

1. INTRODUCTION

Fault diagnosis and condition monitoring have been studied in the recent decade to preventcostly interruptions due to motor faults and recognize faulty conditions as soon as possible[1–7]. Electrical motors are subjected to faults which may redound to secondary faults. Thesources of motor faults may be internal, external or due to environmental conditions. Internalfaults can be classified with reference to their origin.

Internal faults can be classified with their outbreak location: stator or rotor. Commonmachine faults in rotor according to [8] are:

1) Bearing failure;2) Rotor broken bars;3) Rotor body failure;4) Bearing misalignment;5) Rotor misalignment;6) Bearing loss of lubrication;7) Rotor mechanical or thermal unbalanced;

And common faults become apparent in stator as categorized in [8] are:

1) Frame vibration;2) Stator earth faults;3) Damage of insulation;4) Stator turn-to-turn faults;5) Stator phase- to- phase faults;6) Displacement of conductors;7) Failure of electrical connections;

These failures can be detected with several procedures. In this paper, they are discussed bytheir detection method and parameters will be measured to four groups.

2. FAULT DETECTION METHODS

There are several indicators for faulty conditions of rotating electrical machines help us todistinguish machine conditions. In this paper, fault detection methods persuaded by theirfault indicators. So condition monitoring method can be analyzed in four groups aspresented in Fig. 1.


312

Fig. 1. Fault diagnosis methods

2.1 Electrical Analysis

Some of the electrical faulty condition symptoms are motor current signature, voltage, flux,power and so on. Probable faults can be detected by comparison between electrical signalsin healthy and unknown conditions.

Some of the electrical methods are based on signal injection and response analysis. Forinstance, a method based on signal injection with high-frequency proposed in [9] for faultdetection in closed-loop drives, but it’s difficult to implement for many applications due toinvasiveness and hardware limitations.

Akin et al. in [10] reported that the reference frame theory directly added into the main motorcontrol subroutine in DSP program can successfully be applied to real-time fault diagnosis ofelectric machinery systems to find the magnitude and phase quantities of fault signatureseven though in nonideal conditions such as offset, unbalance, etc.

In the rated rotor flux test by applying an ac voltage source across each side of the shaft,high shaft current and yoke flux have been utilized. This induces circulating current betweenthe rotor bars and shaft, and the current or flux of each bar is indirectly monitored using ironfilings/magnetic viewer or a thermal imaging camera. The influence of a cracked or brokenbar or shorted rotor laminations can be observed by this test [11]. These methods are beingdone under standstill condition and don’t seem efficient for online condition monitoring.

An automated technique for monitoring of rotor condition of voltage source inverter-fedinduction machines at standstill has been proposed in [11]. In this algorithm, the motor isexcited with a set of pulsating fields at a number of angular positions for observing thechange in the impedance pattern for broken bar detection. This technique can be performedwithout any extra hardware but it’s still an offline test.


313

2.1.1 Motor current signature analysis (MCSA)

MCSA is one of the most popular approaches since it provides sensor less diagnosis of rotorand bearing problems [11,12,13]. MCSA requires the measurement and manipulation oflengthy steady-state data and an accurate measurement/estimate of the rotor speed forobtaining a reliable and high-resolution assessment but MCSA is not so effective forapplications where the load constantly changes.

The prior MCSA techniques assume stationary and high SNR for signal. The nonstationaryof stator current is accommodated by the commonly used windowing techniques [14]. Thehighly transient and dynamic nature of the induction motor stator current during faultconditions demand analysis through algorithms and techniques fit to analyze nonstationaryand nonlocalized signals, such as wavelet transform or other time-frequency techniques.The availability of the advanced signal processing tools, such as higher order spectrumanalysis [15], high-resolution or subspace methods [16] and wavelet analysis [17,18] haverevolutionized the signal processing for fault detection in electrical motors.

MCSA usually has been attempted looking at fs)21( and fs)21( frequencies, lowersideband (LSB), and upper sideband (USB), which s is slip and f is main frequency [19].The sideband amplitudes are affected by load level and power rating, constructive details,and by manufacturing asymmetries [20].

Because of the vicinity of signal main frequency to produced components and sidebands,broken bar detection may be difficult by this method [21]. Also, this problem exists under lowslip operation. MCSA-based online rotor fault detection is not very effective since the currentregulator masks the fault signatures in the current [22-24]. In addition, online monitoringtechniques can fail if the operating frequency constantly changes due to adjustable speedoperation. In [23,24], spectrum analysis of variable speed controller was proposed for rotorfault detection in field-oriented drives, but the methods can only be applied for a specificcontrol scheme and are strongly influenced by controller parameters [25].

In [19] some new fault indicators for bar-breakage detection are exposed based on thesidebands of phase-current upper harmonics; the ratios

f

fsI

I5

)27( andf

fsI

I7

)25( are examples

of such indicators, and they are independent on load torque and drive inertia. This methodhas low independence with respect to machine parameters and has linear dependence onfault gravity.

Jung et al. in [26] conducted an advanced online diagnosis system using MCSA and madeup of the optimal slip-estimation algorithm, the proper sample selection algorithm, and thefrequency auto search algorithm for more productivity.

In [27] have been compared different fault diagnosis methods like three phase currentvector, the instantaneous torque, and the outer magnetic field. Finally, it’s declared thatMCSA can be the best method for diagnosis the rotor faults.

As a basic tool, various reference-frame-theory-based applications are reported in the recentstudies, like finding deviation in an actual Concordia pattern used to determine the types andmagnitude of faults in drive systems and stator, respectively [28,29], obtaining negative-sequence stator-fault-related indices from the line current [30], and detecting negative-


314

frequency rotor asymmetry signatures at standstill based on complex fault signature vectors[31].

Time-frequency analysis has been investigated vastly in recent years but its complexity andheavy hardware requirements are limitations for simple low-cost drive systems [22].

There are several ways for data comparison in signal processing like Kolmogorov–Smirnov(KS) technique, Plateau algorithm, Holf–Winters (HW) technique and Mark–Burgess (MB)technique. If two time data series or distributions are at a significant variance the KStechnique [32,33], a nonparametric and distribution-free technique [34] is best choice. Theyare being used for comparison motor current signal with reference signal. The referencesignal is motor current signal in healthy condition. The KS parameter is evaluated by takingthe vertical difference between the two data distributions under test into consideration. ThePlateau algorithm is apposite for handling long-term deviations and seems not suitable forcondition monitoring. Holf–Winters (HW) algorithm is a forecasting technique needs aspontaneously event detection procedure, and Mark–Burgess (MB) technique is intended fordetecting real-time changes. The KS technique is the best known of several distribution-freetechniques that test general differences between data distributions. It is more valuable forapplications, which are responsive to data distributions [14].

2.1.1.1 Order tracking method

Similar to vibration analysis in nonstationary condition or in variable speed motors instead oftracking absolute frequency, frequencies can be explained by multiple of a base frequencythat is usually power source frequency. For instance this method in [35] used for detectioninter-turn in Permanent Magnet Synchronous Motor (PMSM). In [35] by applying a Vold-Kalman Filter (VKF) [36] tried to use order tracking method for selected voltage and currentharmonics and detect inter-turn in PMSM. Vold-Kalman Filter Order tracking (VKF-OT)beneficiary is that allows extracting both the amplitude and phase of the analyzed orders ateach time instant directly from the original data. Furthermore, its tracking performance doesnot depend on the slew rate (rotational speed rate of change) [35] and make order trackingon noisy signal easy.

2.1.1.2 Time and frequency domain analysis

There are some restrictions of the Fourier transform, for example it cannot be used for nonperiodic or nonstationary signals; otherwise, the resulting FFT spectrum will make littlephysical sense [17,37,38].

However, for machinery operating under unsteady conditions, because of variation in therotating speed and operating load, even if the machine is in the normal state, the spectrumof the vibration signal is always altering in sampling time. When a nonstationary signal istransformed into the frequency domain, most of the information about the transientcomponents of the signal will be lost [39], hence, a hybrid method has been proposed in[40].

Time-frequency analysis [41] methods can simultaneously generate both time and frequencyinformation from a signal. Therefore, in later studies, time-frequency analysis methods arewidely used to detect faults since they can determine not only the time of occurrence butalso the frequency ranges of the location [42]. Time-frequency methods mostly use invibration analysis and MCSA. There are several time-frequency analysis methods, such as


315

the Short-Time Fourier Transform (STFT), Wavelet Analysis (WA), and the Wigner-VilleDistribution (WVD), which may be used for condition monitoring of rotating machinery intransient and unsteady operating conditions. Those time-frequency techniques have beenapplied to fault diagnosis and condition monitoring in practical plant machinery [18,43,44].Also Hilbert transform and Zhao–Atlas–Marks distribution in [45] applied to fault diagnosis ofmotors in nonstationary conditions but this method is not as common as prior methods.

Misalignment detection using STFT and WA signal processing techniques is shown in Fig. 2[25].

(a) (b)

Fig. 2. Misalignment detection using STFT and the wavelet technique: (a) STFT (b)STFT and wavelet technique [3]

In the field of machinery fault monitoring, Wavelet Analysis (WA) has been used widely inthe diagnosis of rolling bearings, gearbox and compressors. This technique also has beenused for feature extraction and noise cancellation of the various signals [18,43-46].

In [18,43,47], a fault diagnostic technique for rotating machinery is investigated based ondiscrete wavelet transform. In Reference [48] a time-averaged WA according to Morletcontinues wavelet used for fault diagnosis of a gear set. Also, reference [49] presents acombination of Continuous Wavelet Transform (CWT) and Kolmogorov-Smirnov test for faultdetection of the bearings and gear box in transient conditions. In [46,50] CWT is used forextract the features of roller bearing fault signals. Reference [51] used CWT for fault signaldiagnosis in an internal combustion engine.

In [52], the application of the Wigner-Ville distribution is reported to detect a broken tooth in aspur gear. Reference [53] shows that the WVD can be applied to the description of machineconditions and it is an effective method in machinery fault diagnosis. Reference [44] appliesa PWVD to identifying the influence of the fluctuating load conditions for gearbox. A DigitalSignal Processing (DSP) implementation is presented in [54] to detect mechanical loadfaults in induction motors during speed transients based on WVD and stator current analysis.

2.1.2 Flux monitoring

Magnetic flux can be a fault indicator and monitored both inside the machine (search coils)or outside (axial coils). Coil installation and noisy spectra are the main difficulties [19]. Oneof the most applications of this algorithm is fault detection in rotor cage. The estimated rotorflux in [24] suggested for the diagnosis of rotor faults in vector-controlled drives. In [84]


316

Dorell et al. showed a relation between air gap eccentricity and air gap flux and vibrationsignals.

Cruz et al. in [55] presented an algorithm for diagnosis of rotor faults which starts with themeasurement of the amplitude of the rotor flux oscillations. It’s showed that the ratiobetween dsi and the average value of qsi , current changes in d and q axis respectively,gives the degree of asymmetry of the motor or the number of adjacent broken bars, if thetotal number of rotor bars is known. But this algorithm needs some additional modules forcalculating the current average values and tracks the amplitude of currents.

2.1.3 Motor power monitoring

Motor power signature analysis is focused on the detection of double-slip frequenciespresent in the electric input power spectrum [56] similar to MCSA. These harmonics areevaluated with respect to the average power (dc component), thus obtaining some faultseverity factors. In addition, this method needs to acquire both currents and voltages. Alsothe dependence on the drive inertia is another limitation of this fault indicator [57]. Bellini etal. in [57] tried to detect rotor broken bar by this approach.

2.1.4 Partial discharge (PD) monitoring

This test mainly used in high voltage motors and generator stator windings. By using PartialDischarge Analyzer (PDA) sensors placed within the winding or at the winding terminals,stator winding PD pulses will separate from electrical interference (usually harmless) basedon pulse arrival time or pulse shape and easily can be detected [58]. PD is a symptom ofmany stator winding insulation failure mechanisms. IEEE 1434-2000 reviews all types of PDmeasurement methods used in rotating machines [59].

There are several discharge monitoring techniques. Among these methods RF couplingmethod, capacitive coupling method and broad-band RF method [60] are more known. ARadio Frequency Current Transformer (RFCT) installed on neutral point of winding candetect Radio Interference Frequency Intensity (RIFI) caused by PD. Arcs occurred at anylocation cause RF current flow into the neutral point because of its low potential. The RIFImeter had a narrow bandwidth of about 10 kHz centered at 1MHz [60]. By using afrequency-based method with low power hardware, it is possible to take advantage of the RFtechnique without the need for wideband signal capture and its associated overheads [61].

Second method use specialized pulse height analyzer with bandwidth 80 MHz. In thisapproach connection to the winding is made through coupling capacitors at the machine lineterminals [60]. Initially, the capacitors were connected to the machine during an outage, butlatterly described how the capacitors could be permanently built into the phase rings of themachine and the measurements can be made without service interruption. In [62] showedthat the pulse has a rise time (defined as 10%–90% of peak) of 4 ns and the frequencycontent of this pulse extends to over 100 MHz, thus, an 80-pF capacitor installed on high-voltage machine terminals can be used as the coupling device.

It has been shown that serious PD, sparking or arcing, has faster rise-times than thebackground corona and PD activity, and therefore produce a much higher bandwidth ofelectromagnetic energy, up to 350 MHz. If this energy is detected, at as high a frequency aspossible, the ratio of damaging discharge signal to background noise is increased.


317

Frequencies above 0.4 MHz do not propagate from the discharge place along the winding,as with the lower frequency techniques, but by radiation from the winding [60]. This radiationcan be detected by an RF aerial located inside the enclosure of the machine or outside,close to an aperture in it and it is basic concepts of broad-band RF monitoring method.

2.1.5 Voltage spectrum analysis

The Growler test and rated rotor flux test with high current ac excitation are anothercommonly used offline tests for rotor testing [63-67]. A Growler is an electrical device usedfor testing insulation of a motor for shorted coils with an iron core and excited by AC currentfor detection insulation problem.

The method consists of inserting an auxiliary small winding which is a coil “sneak’’ thatforms an angle 0 with the A stator phase as shown in Fig. 3 [68]. This coil has noconductive contact with the other phases but it is mutually coupled with all the other circuitson both the stator and rotor sides [69].

Fig. 3. Auxiliary winding emplacement [69]

Mirimani et al. in [70] investigated the effect of static eccentricity on the back EMF of an AxialFlux Permanent magnet (AFPM) through 3D-FEM (Finite Element Method) as shown in Fig.4 [68]. The back EMF of the four coils of one phase is obtained to propose a suitablecriterion for precise eccentricity fault detection.

Fig. 4. 3D-FEM model of the axial flux permanent magnet motor [68]


318

In the case of a healthy motor the auxiliary winding voltage Park components spectracontain one peak at the motor main supply frequency. The Lissajous curve is an ellipse asshown in Fig. 5 [71]. In the different cases of voltage unbalances, the Lissajous curves arealso ellipses that have different angles as shown in Fig. 6 [71]. In comparison with damagedand non defected motor, the value of their superior and inferior radiuses will increase [68].

It is also well known that the effects of stator winding inter-turn faults may be detected bymonitoring the Zero-Sequence Voltage Component (ZSVC) [72,73]. This method benefit isthat it’s separate from motor drive against some other methods like MCSA, but it needs toaccess to stator winding neural point. In [35] attempted to detect inter-turn fault in PMSM byfirst harmonic amplitude of ZSVC and stator currents third harmonic. Briz et al. [74] usedvoltage and current zero-sequence components for recognition of faults in inductionmachine.

Fig. 5. Park’s Currents Vector of a healthy motor [71]

Fig. 6. Park’s currents vector for a motor with coils in shortcut [71]


319

2.2 Mechanical Analysis

There are several mechanical symptoms for faulty condition of electrical machine, such as:vibration, noise, torque and so on.

2.2.1 Vibration monitoring

As almost 80 percent of common rotating equipments problems are related to misalignmentand unbalance, vibration analysis is an important tool that can be used to eliminate recurringproblems [75,76]. In many cases, the overall vibration level of the machine is sufficient todiagnose mechanical failures [77,78], but in [2] showed that this is not an efficient method forall faults. In [79] showed that the electromagnetic force is the most sensitive indicator of airgap eccentricity. Therefore identifiable signatures should be found in the vibration pattern ofrotating electrical machines. The only drawback of this indicator is its low accessibility.Nevertheless, since vibrations are the consequences of the forces on the machine structure,identifiable signatures should be found in the vibration pattern. The measured vibration andassociated current harmonics are closely correlated [14].

Literature survey [80-83] shows that most of the bearing fault diagnoses are based onvibration analyses like wavelet transform and Hilbert–Huang transforms or current-basedanalysis.

In [84] illustrated how eccentricity faults can be identified from vibration analysis usingcondition monitoring techniques.

The overall RMS of vibration can be calculated by different definition based on the spectrumin frequency domain across all of the effective frequency range, i.e., from DC to maximumanalysis frequency range. One of the suggested formulas is [85]:

BW

fpoweroverallRMS

sf

45.0

0)(

(1)

In above equation, BW is noise power bandwidth of window, f is analysis frequency band

and sf is sampling frequency band.

Another special frequency analysis is Cepstrum that defined:

221 )})}({{log()( tfFFC (2)

This can be used for examining behavior of gearboxes [21].

2.2.1.1 Frequency-domain analysis

The most common tools of vibration monitoring in industrial plants is frequency analysis.Finley et al. [86] compiled a resume table with a comprehensive list of electrically and


320

mechanically induced components in the vibration pattern. Their analysis is based onanalytical formulas.

In [87], a strategy presented based on monitoring slot passing frequencies in high frequencyvibration components. Their presented analysis was based on rotating wave approachwhereby the magnetic flux waves in the air gap are taken as the product of permeance andMagneto Motive Force (MMF).

Vibration pattern for the healthy motor and with dynamic eccentricity has been compared in[88] as shown in Fig. 7. In paper [88] has been showed that the low frequency componentsof vibration (measured by accelerometers fixed on the outer casing of motor) can be used assignatures for the detection of eccentricity in induction motors.

2.2.1.2 Order tracking methods

The advantages of order tracking over the other vibration techniques mainly lie in analyzingnon stationery noise and vibrations which will vary in frequency and amplitude with therotation of a reference shaft. The analysis of non stationery conditions needs additionalinformation, as compared to steady state conditions, for an accurate result to be obtained.Order domain analysis relates the vibration signal to the rotating speed of the shaft, insteadof an absolute frequency base [21].

Fig. 7. Vibration pattern for healthy motor (top) and with 37% dynamic eccentricity(bottom), 1.9% and motor fed at 100Hz in both cases [89]

2.2.2 Noise monitoring

Measuring and analyzing the acoustic noise spectrum [90] is another method of conditionmonitoring in rotating electrical machinery which require special consideration. Acousticnoise emitted from air gap can be an indicator of probably eccentricity in induction motor.But, the application of noise measurement in a noisy environment like a plant is not soefficient. In [89] an approach for air gap eccentricity detection presented and a test carriedout in an anechoic chamber. Slot harmonics in the acoustic noise spectra were introduced


321

as an indicator of static eccentricity. Li and He [1] used Hilbert-Huang Transform (HHT) foranalyzing nonstationary noise signals incorporates a threshold-based denoising technique toincrease the SNR for health monitoring in electrical machines.

Reference [91] examines whether acoustic signal can be used effectively to detect thevarious local faults in gearboxes using the smoothed Pseudo Winger-Ville Distribution(PWVD).

Scanlon et al. [92] showed that by extraction hide information of acoustic noise signal canpredict machinery resident life time.

Defects in the roller element bearings cause particular frequencies to be excited. Thesefrequencies can be detected in acoustic noise spectrum. In [93], an automated approach todegradation analysis is proposed that uses the acoustic noise signal from a rotating machineto determine the remaining useful life of the machines.

2.2.3 Torque monitoring

By comparison between the estimated torque from the model and measured torque candetect some faults in electrical motors, so it’s necessary to have a good model and analgorithm to be aware of air gap real torque. The electromagnetic torque estimation hasbeen commonly used in electrical drives to control the torque and the rotor speed of ACelectrical machines. So, it is needed to compute stator flux or rotor flux exactly in which theaccuracy and the robustness are directly related to electrical machine parameters [94]. Inaddition, the flux estimation needs to have knowledge about only two parameters of thesethree parameters: stator phase voltages, currents, and the rotor speed by using anappropriate model [95].

In reference [96] torque estimation beside torsional vibration analysis used for gearbox faultdetection in traction system and by measuring the torque their work has been validated.

Guzinski et al. in [97] for identification problems related to transmission system in HighSpeed Train (HST) used the load torque observer without adding any additional sensors.The presented observer system was able to detect the meshing frequency of the test benchwhich has very small amplitude in the tested healthy gear.

From the input terminals, the instantaneous power includes the charging and dischargingenergy in the windings. Therefore, the instantaneous power cannot represent theinstantaneous torque. From the output terminals, the rotor, shaft and the mechanical load ofa rotating machine constitute a torsional spring system. This torsional spring system has itsown natural frequency [98]. The attenuation of the components of the air gap torquetransmitted through the torsional spring system is different for different harmonic orders oftorque components [99,100].

The locked-rotor torque and breakdown torque will decrease in unbalanced voltage situation.If the unbalanced voltage was extremely severe, the torque might not be adequate for theapplication although the full-load speed is reduced slightly when the motor operates withunbalanced voltages [101] and it can be an indicator of unbalance voltage condition.


322

2.3 Chemical Indicators

Insulation degradation can be monitored chemically by the presence of special matter in thecoolant gas or by detection some particular gases such as ozone, carbon monoxide or evenmore complex hydrocarbons, like acetylene and ethylene [60]. Electrical discharge activity,heat and some other electrical and mechanical faults may lead to insulation degradation.The product materials can be gas, liquid or solid. Each of them needs a particular detectionmethod.

An ion chamber was designed in [102] to detect the products of heated insulation and it wasapplied to a large turbo generator.

The metal wear debris in oil can be classified ferromagnetic wear debris andunferromagnetic wear debris. When wear debris is in the coil of inductive wear debrissensor, the magnetic field distribution of the coil is changed, and then the equivalentinductance of the coil was changed. This technique for metal wear debris in oil is anoncontacting and quick method and can be off-line and on-line [103].

In addition oil particle can be detected for fault diagnosis. With modern diagnostic tools, oilanalysis is used to monitor the condition of equipment as well as condition of a lubricant.Various faults such as misalignment, unbalance, overload or accelerated heating conditionmay lead to wearing in electrical machinery. The different types of wear are: abrasive wear,adhesive wear, cavitations, corrosive wear, cutting wear, fatigue wear and sliding wear [75].Some types of oil analyses are: viscosity, solids content, water content, total acid number,total base number and flash point [75].

As mentioned, wear particles are the prime indicators of the machine’s health. There aremany techniques to evaluate the type and concentration of such particles. The techniquesinclude: spectrometric analysis, infrared analysis, X-ray fluorescence (XRF) spectroscopy,particle counting, direct reading ferrography and analytical ferrography [75].

2.3.1 Spectrometric analysis

This is one of the main techniques that typically reported in PPM (Parts Per Million). Thistechnique generally monitors the smaller particles and large wear metal particles present inthe oil will not be detected.

For larger wear particles, there are available techniques such as: acid digestion method,microwave digestion method, direct read (DR) ferrography and Rotrode filter spectroscopy(RFS).

2.3.2 Infrared analysis

Specific groups of atoms called functional groups by this method can be detected. Anappropriate wavelength is directed at the sample being analyzed, and the amount of energyabsorbed by the sample is measured. The amount of absorbed energy is an indication of theextent of presence for that particular functional group in the sample. It is hence possible toquantify the results. This analysis was first introduced in 1979. After several years a newmethod extracted from this analysis named Fourier Transform-Infrared Analysis (FT-IR). Bythis technique, a beam of light is focused through a film of used oil and the wavelengths are


323

then compared to light transmitted through new oil of the same type. The differences inreadings provide information with respect to the degradation of the used oil [75].

2.3.3 Wear particle analysis (WPA) or ferrography

Ferrography or WPA utilizes microscopic analysis to evaluate the particles type, shape, sizeand quantity. The components specifications allow a process of elimination in which theabnormal wear can be identified. This analysis is used in two ways: A routine monitoring andtrending of the solid contents, Observing and analyzing the type of wears [75,104].

2.3.4 XRF (X-ray fluorescence) spectroscopy

The XRF spectroscopy entails the excitation of electrons from their orbits. This leads toemission of UV rays with characteristic frequencies, which can be analyzed. During Rotrodeatomic emission spectroscopy, an electrical discharge produces plasma, causing thermalemission. When the atoms return to the normal state, the excess energy is emitted as light.Each element emits light at different frequencies on the electromagnetic spectrum. Theamount of light emitted at a given frequency corresponds to the concentration of the elementpresent in the sample. Also atoms can be excited by bombardment of X-rays [75].

2.3.5 Image processing

The image processing and computer vision system reveals more information in the form ofquantitative data not revealed by the human eye. This technique is used to collectquantitative information from wear particle images. Image analysis system is developed toprocess and store the information of particle shape and edge detail features. In [105]particles have been defined as regular, irregular, circular and elongated. So, an imageprocessing technique is applied for analyzing wear debris.

2.4 Thermal Monitoring

Due to thermal limitation of various parts of rotating electrical machines such as insulations,coil and so on, it’s necessary to have a good idea about machine parts temperature.Thermal monitoring for electrical machines has two aspects, measuring the temperature andthermal modeling, which each one of them has been illustrated shortly.

Also recently a new wireless sensor for bearing temperature monitoring presented [106].This sensor is a combination of a ring-shaped permanent magnet and a Hall Effect sensorthat detect variation in magnetic field because of growing in temperature.

2.4.1 Temperature measurement

There are three main approaches for temperature measurement in electrical machines: 1)Measuring local point temperatures by embedded temperature detectors (ETD) or resistancetemperature detectors (RTD); 2) Using thermal images, fed with suitable variables, tomonitor the temperature of the perceived hottest spot in the machine; 3) Measuringdistributed temperatures of the machine or bulk temperatures of the coolant fluid [60].

These demonstrate the fundamental difficulty of temperature monitoring; the conflictbetween easily made point measurements, which give only local information, and bulk


324

measurements that are more difficult and run the risk of overlooking local hot-spots.Choosing location of settling detectors requires careful consideration during specification.Bulk measurement can be found from the measurement of the internal and external coolanttemperature rises, obtained from thermocouples located.

Milic and Srechovic in [107] presented a new non-contact measurement system for hotspotand bearing fault detection in railway traction system (RTS).

Of course, due to rotating parts in electrical motors, these methods are not efficient andthermal modeling is inevitable.

2.4.2 Thermal modeling

Generally, thermal models of electric machines are classified into two categories [98,108]:

1) Finite Element Analysis (FEA) based model2) Lumped Parameter (LP) thermal model

Finite Element Method (FEM) or Finite Difference Method (FDM) tools have traditionallybeen used to model the thermal performance of electric machines. Their applications havebeen limited only to small sectors of the stator and rotor and have not shown full-scalesimulation for motors with complicated geometry. The accuracy of model is generallydependent on the number of thermally homogenous bodies used in model [109,110]. By thiswork, researcher may simplify the complicated geometry and shorten computational time forconstructing elements and calculating large system matrices.

On the other hand lumped parameter equivalent thermal circuit is easy to solve and gives agood overall view of the temperature rise in different parts of the machine without muchcomputational time [111]. Chowdhury claimed that the lumped parameter thermal equivalentcircuit proposed in [112] is easy to visualize as all the parameters are directly derived fromthe machine geometry. Boglietti et al. [108] compared the LP and FEA for thermal modelingof electrical machines.

There are two ways for extraction parameters of lumped parameter model. The first one isby using comprehensive knowledge of the motors, physical dimensions and constructionmaterials. The second one is to identify the parameters from extensive temperaturemeasurement at different locations in the motor explained in previous session. Even thoughan electric machine is made up of various materials that have different characteristics, themachine can be assumed to consist of several thermally homogenous lumped bodies [98].For example, a simplified model of an induction model and a PMSM consisting of twolumped thermal bodies are presented in [113,114]. Likewise in [115], Milanfar and Langdeveloped a thermal model of electric machine to estimate the temperature of the motor andto identify faults like turn-to-turn faults and bearing faults.

A time-domain lumped thermal model of an induction motor obtained in [116]. Thetemperature distribution and the energy destruction are shown in Fig. 8.

Nategh et al. in [117] presented a lumped parameter thermal model for a permanent-magnetassisted synchronous reluctance machine (PMaSRM) developed for propulsion in a hybridelectric vehicle. They divided the stator slot into a number of elliptical copper andimpregnation layers and modeled stator winding by some approximation.


325

(a)

(b)

Fig. 8. Thermal model of an induction motor in the flow loop 3 h after startup. Thetemperature distribution is shown in (a), and the energy destruction is shown in (b)

[116]

Jankowski et al. [116] described the development of a time-dependent lumped-parameterthermal model of an induction motor, and showed that how this thermal model can be usedto minimize the internal temperature during operation.

Kolondzovski et al. in [118] discussed about thermal issues of different types of electricmotors and different rotor types. Similarly, EL-Refaie et al. in [119] presented multibarrierinterior permanent magnet machines lumped parameter model.


326

Idoughy et al. [120] proved that the analytical techniques may risk underestimating thehotspot winding temperature, especially when the fill factor is below 0.3. In addition, thetemperature variation in the axial direction is not considered and hotspot temperatures oftenarise in the end windings.

In [121,122] it’s claimed that they can calculate rotor and stator respectively under thesteady state and transient steady by off-line experiment and their model can respond tochanges in the cooling conditions. However, their models are generally sensitive to unknownmachine parameters and their variation. Also, by DC signal injection thermal parameter ofelectrical machines components can be achieved [123,124]. This method applied forinduction motors fed by closed-loop inverter drives in [125].

3. MODEL BASED & AI-BASED METHODS

A model-based fault monitoring method presented in [126] for variable speed drives withoutfrequency analysis. Nowadays, AI-based which use fuzzy logic, neural network, particleswarm optimization [127] and so on are so popular for researchers. Some of them areexplained in this paper.

3.1 Artificial Neural Network

Nejjari et al. in [128] used learning Park’s vector pattern based on artificial neural network todiscern healthy and faulty patterns. Also, Wang et al. in [129] used combination of these twoalgorithms for condition monitoring of rolling bearings.

Tag Eldin et al. [130] used Artificial Neural Network and applied result of the RMSmeasurement of stator voltages, currents and motor speed to train a neural network tomonitor and diagnosis external motor faults.

Asiri [131] decided to detect six different types of PD using neural networks and classifydifferent types of PD according to the location of PD activity.

3.2 Fuzzy Logic

The fuzzy logic tool provides a technique to deal with imprecision and recently attractedresearchers attention for different applications like fault diagnosis. The utility of fuzzy setslies in their ability to model uncertain and vague data. Fuzziness in a fuzzy set ischaracterized by its membership functions [132].

An extraction method based on the Relative Crossing Information (RCI) in [133] proposed forcondition monitoring of a machine under the variable rotating speed, by which theinstantaneous feature spectrum can be automatically extracted from the time-frequencydistribution of the fault signal. The performance of this approach is evaluated using threetime-frequency techniques, namely STFT, WA, PWVD and finally using a sequential fuzzydiagnosis method.

Reference [134] claimed that using fuzzy sets and uncertainty phenomena with possibilitytheory may help in fault diagnosis of satellite applications. A combination of neural networkand fuzzy logic used in [129] for condition monitoring of rolling bearings. Also, [135]propounds an intelligent condition diagnosis method for rotating machinery developed using


327

least squares mapping (LSM) and a fuzzy neural network. In [133], possibility theory is alsoapplied to combine with PWVD technique for fault diagnosis.

4. CONCLUSIONS

Condition monitoring methods for rotating electrical machines have been surveyed in fourgroups. These groups consisted of: electrical analysis, mechanical analysis, chemicalanalysis and thermal analysis. In each group, there are several symptoms that faultycondition in machines can be detected by them.

Methods based on signal injection seem profit for fault detection in closed-loop drives, butit’s difficult to implement for many applications due to invasiveness and hardware limitations.MCSA, the most popular technique, provides sensor less diagnosis of some motor problemsbut it’s not so effective for applications where the load constantly changes. Time-frequencyanalysis has been investigated vastly in recent years but its complexity and heavy hardwarerequirements are limitations for simple low-cost drive systems.

Motor power analysis because of need to both currents and voltages simultaneously anddependence on the drive inertia has some limitation. PD monitoring mainly used in highvoltage motors and generator stator windings. Most of recurring problems in rotatingmachinery like misalignments can be detected by vibration analysis. The measured vibrationand associated current harmonics are closely correlated. By detection ozone, carbonmonoxide and others in the coolant gas or oil analysis, some faults like insulationdegradation can be detected easily. Also thermal measurement and thermal modeling areintroduced as efficient tools for motors condition monitoring. Finally, AI- based algorithmscombined of one or more explained methods were studied.

Besides these methods and algorithms, nowadays web-based monitoring approaches areinteresting. They are using one or more of these mentioned procedures in softwares likeLabVIEW, as you see in [136] and shown in Fig. 9.

Fig. 9. Diagnosis session panel with decision method menu presented in [136]


328

COMPETING INTERESTS

Authors have declared that no competing interests exist.

REFERENCES

1. Ruoyu Li, David He. Rotational Machine Health Monitoring and Fault Detection UsingEMD-Based Acoustic Emission Feature Quantification. Instrumentation andMeasurement IEEE Transactions. 2012;61(4):990-1001.

2. Mortazavizadeh SA, Vahedi A, Zohouri AA. Detection of Stator Winding Inter-turnShort Circuit In Induction Motor Using Vibration Specified Harmonic Amplitude , 2ndInternational Conf. on Acoustics & Vibration, Sharif Univ. of Technology, Tehran, Iran.Dec. 2012.

3. Hachemi Benbouzid El, Mohamed. A review of induction motors signature analysis asa medium for faults detection. Industrial Electronics, IEEE Transactions.2000;47(5):984-993.

4. Subhasis N, Raj Mohan Bharadwaj, Hamid A. Toliyat. Performance analysis of athree-phase induction motor under mixed eccentricity condition. Energy Conversion,IEEE Transactions. 2002;17(3):392-399.

5. Dorrell, David G, William T. Thomson, and Steven Roach. Analysis of airgap flux,current, and vibration signals as a function of the combination of static and dynamicairgap eccentricity in 3-phase induction motors. Industry Applications, IEEETransactions. 1997;33(1):24-34.

6. Devaney, Michael J, Levent Eren. Detecting motor bearing faults. Instrumentation &Measurement Magazine. IEEE. 2004;7(4):30-50.

7. Riley, Caryn M., et al. Stator current harmonics and their causal vibrations: apreliminary investigation of sensorless vibration monitoring applications. IndustryApplications, IEEE Transactions. 1999;35(1):94-99.

8. Singh GK, Ahmed Saleh Al Kazzaz. Induction machine drive condition monitoring anddiagnostic research—a survey. Electric Power Systems Research. 2003;64(2):145-158.

9. Briz, Fernando, et al. Online diagnostics in inverter-fed induction machines using high-frequency signal injection. Industry Applications. IEEE Transactions.(2004);40(4):1153-1161.

10. Akin, Bilal, et al. A simple real-time fault signature monitoring tool for motor-drive-embedded fault diagnosis systems. Industrial Electronics, IEEE Transactions.2011;58(5):1990-2001.

11. Kim, Byunghwan, et al. Automated detection of rotor faults for inverter-fed inductionmachines under standstill conditions. Industry Applications, IEEE Transactions.2011;47(1):55-64.

12. Bouchikhi E El, Choqueuse V, Benbouzid ME. Current frequency spectral subtractionand its contribution to induction machines’ bearings condition monitoring. IEEE Trans.on Energy Conversion. 2013;28(1):135-144.

13. Filippetti F, Bellini A, Capolino G. Condition monitoring and diagnosis of rotor faults ininduction machines: state of art and future perspectives. Electrical Machines DesignControl and Diagnosis (WEMDCD), IEEE Workshop; 2013.


329

14. Gupta RA, Wadhwani AK, Kapoor SR. Early estimation of faults in induction motorsusing symbolic dynamic-based analysis of stator current samples. Energy Conversion,IEEE Transactions. 2011;26(1):102-114.

15. Arthur, Neil, Jim Penman. Induction machine condition monitoring with higher orderspectra. Industrial Electronics, IEEE Transactions. 2000;47(5):1031-1041.

16. Bellini, Alberto, Giovanni Franceschini, Carla Tassoni. Monitoring of inductionmachines by maximum covariance method for frequency tracking. IndustryApplications, IEEE Transactions. 2006;42(1):69-78.

17. Peng ZK, Peter W, Tse, Chu FL. A comparison study of improved Hilbert–Huangtransform and wavelet transform: application to fault diagnosis for rolling bearing.Mechanical systems and signal processing. 2005;19(5):974-988.

18. Purushotham V, Narayanan S, Suryanarayana AN Prasad. Multi-fault diagnosis ofrolling bearing elements using wavelet analysis and hidden Markov model based faultrecognition. NDT & E International. 2005;38(8):654-664.

19. Bruzzese, Claudio. Analysis and application of particular current signatures(symptoms) for cage monitoring in nonsinusoidally fed motors with high rejection todrive load, inertia, and frequency variations. Industrial Electronics, IEEE Transactions.2008;55(12):4137-4155.

20. Bellini, Alberto, et al. On-field experience with online diagnosis of large inductionmotors cage failures using MCSA. Industry Applications, IEEE Transactions.2002;38(4):1045-1053.

21. Wang, Ke Sheng. Vibration monitoring on electrical machine using Vold-Kalman filterorder tracking. Diss. University of Pretoria; 2008.

22. Bellini, Alberto, et al. Advances in diagnostic techniques for induction machines.Industrial Electronics, IEEE Transactions. 2008;55(12):4109-4126.

23. Bellini, Alberto, et al. Closed-loop control impact on the diagnosis of induction motorsfaults. Industry Applications, IEEE Transactions. 2000;36(5):1318-1329.

24. Cruz SMA, Cardoso AJM. Diagnosis of rotor faults in closed-loop induction motordrives. Industry Applications Conference, 2006. 41st IAS Annual Meeting. ConferenceRecord of the 2006 IEEE. Vol. 5. IEEE; 2006.

25. Cruz, Sérgio MA, et al. A new model-based technique for the diagnosis of rotor faultsin RFOC induction motor drives. Industrial Electronics, IEEE Transactions.2008;55(12):4218-4228.

26. Jung, Jee-Hoon, Jong-Jae Lee, Bong-Hwan Kwon. Online diagnosis of inductionmotors using MCSA. Industrial Electronics, IEEE Transactions. 2006;53(6):1842-1852.

27. Szabo L, et al. An Overview on Induction Machine's Diagnosis Methods. Journal ofComputer Science and Control Systems; 2008.

28. Gilreath, Phil, Brij N Singh. A new centroid based fault detection method for 3-phaseinverter-fed induction motors. Power Electronics Specialists Conference, 2005.PESC'05. IEEE 36th. IEEE; 2005.

29. Xiaoping Tu, et al. Modeling and real-time simulation of internal faults in synchronousgenerators with parallel-connected windings. Industrial Electronics, IEEETransactions. 2007;54(3):1400-1409.

30. Cruz, Sergio MA, Marques Cardoso AJ. Multiple reference frames theory: A newmethod for the diagnosis of stator faults in three-phase induction motors. EnergyConversion, IEEE Transactions. 2005;20(3):611-619.

31. Akin, Bilal, et al. DSP-based sensorless electric motor fault-diagnosis tools for electricand hybrid electric vehicle powertrain applications. Vehicular Technology, IEEETransactions. 2009;58(6):2679-2688.


330

32. Tewari, Diwakar. Symbolic dynamic models for highly varying power system loads.Diss. Arizona State University; 2002.

33. Les Cottrell R, et al. Evaluation of techniques to detect significant networkperformance problems using End-to-End active network measurements. NetworkOperations and Management Symposium, 2006. NOMS 2006. 10th IEEE/IFIP. IEEE;2006.

34. Kirman TW. Kolmogorov-Smirnov test at available: http:/www.physics.csbsju.edu/stats/ks-test.html (Tutorial); 1996.

35. Urresty J, Riba J, Luis Romeral. Diagnosis of Inter-Turn Faults in PMSMs Operatingunder Non-Stationary Conditions by applying Order Tracking Filtering. PowerElectronics, IEEE Transactions. 2013;28(1):507-515.

36. Vold, Hsvard. High Resolution Order Tracking at Extreme Slew Rates, Using KalmanTracking Filters. Shock Vib. 1995;2(6):507-515.

37. Peng ZK, Chu FL. Application of the wavelet transform in machine conditionmonitoring and fault diagnostics: a review with bibliography. Mechanical systems andsignal processing. 2004;18(2):199-221.

38. Mitoma, Tetsuro, Huaqing Wang, Peng Chen. Fault diagnosis and conditionsurveillance for plant rotating machinery using partially-linearized neural network.Computers & Industrial Engineering. 2008;55(4):783-794.

39. Douglas, H., Pragasen Pillay, and A. K. Ziarani. Broken rotor bar detection in inductionmachines with transient operating speeds. Energy Conversion, IEEE Transactions.2005;20(1):135-141.

40. Lebaroud, Abdesselam, Guy Clerc. Classification of induction machine faults byoptimal time–frequency representations. Industrial Electronics, IEEE Transactions.2008;55(12):4290-4298.

41. Cohen L. Time-Frequency Analysis. Englewood Cliffs, NJ: Prentice- Hall; 1995.42. Wang, Huaqing, Peng Chen. Fuzzy diagnosis method for rotating machinery in

variable rotating speed. Sensors Journal. IEEE. 2011;11(1):23-34.43. Jian-Da Wu, Chiu-Hong Liu. Investigation of engine fault diagnosis using discrete

wavelet transform and neural network. Expert Systems with Applications.2008;35(3):1200-1213.

44. Stander CJ, Heyns PS, Schoombie W. Using vibration monitoring for local faultdetection on gears operating under fluctuating load conditions. Mechanical Systemsand Signal Processing. 2002;16(6):1005-1024.

45. Rajagopalan S, et al. Non-stationary motor fault detection using recent quadratic time-frequency representations. Industry Applications Conference, 2006. 41st IAS AnnualMeeting. Conference Record of the 2006 IEEE. Vol. 5. IEEE; 2006.

46. Junsheng, Cheng, Yu Dejie, Yang Yu. Application of an impulse response wavelet tofault diagnosis of rolling bearings. Mechanical Systems and Signal Processing. 2007;21(2):920-929.

47. Jian-Da Wu, Chuang-Chin Hsu. Fault gear identification using vibration signal withdiscrete wavelet transform technique and fuzzy–logic inference. Expert systems withapplications. 2009;36(2):3785-3794.

48. Zheng H, Li Z, Chen X. Gear fault diagnosis based on continuous wavelet transform.Mechanical systems and signal processing. 2002;16(2):447-457.

49. Zhu ZK, et al. Detection of signal transients based on wavelet and statistics formachine fault diagnosis. Mechanical Systems and Signal Processing.2009;23(4):1076-1097.


331

50. Tse Peter W, Peng YH, Richard Yam. Wavelet analysis and envelope detection forrolling element bearing fault diagnosis-their effectiveness and flexibilities. Journal ofVibration and Acoustics. 2001;123(3):303-310.

51. Jian-Da Wu, Jien-Chen Chen. Continuous wavelet transform technique for fault signaldiagnosis of internal combustion engines. NDT & E International. 2006;39(4):304-311.

52. Staszewski WJ, Worden K, Tomlinson GR. Time–frequency analysis in gearbox faultdetection using the Wigner–Ville distribution and pattern recognition. Mechanicalsystems and signal processing. 1997;11(5):673-692.

53. Meng, Qingfeng, Liangsheng Qu. Rotating machinery fault diagnosis using Wignerdistribution. Mechanical Systems and Signal Processing. 1991;5(3):155-166.

54. Blodt, Martin, et al. On-line monitoring of mechanical faults in variable-speed inductionmotor drives using the Wigner distribution. Industrial Electronics, IEEE Transactions.2008;55(2):522-533.

55. Cruz SMA, et al. Diagnosis of rotor faults in traction drives for railway applications.Electrical Machines, 2008. ICEM 2008. 18th International Conference. IEEE; 2008.

56. Legowski, Stanislaw F, Sadrul Ula AHM, Andrzej M, Trzynadlowski. Instantaneouspower as a medium for the signature analysis of induction motors. IndustryApplications, IEEE Transactions. 1996;32(4):904-909.

57. Bellini, Alberto, et al. Quantitative evaluation of induction motor broken bars by meansof electrical signature analysis. Industry Applications, IEEE Transactions.(2001;37(5):1248-1255.

58. Stone GC, et al. Development of automatic, continuous partial discharge monitoringsystems to detect motor and generator partial discharges. Electric Machines andDrives Conference Record, 1997. IEEE International. IEEE; 1997.

59. IEEE Trial Use Guide to the Measurement of Partial Discharges in RotatingMachinery. IEEE; 1434-2000.

60. Tavner PJ. Review of condition monitoring of rotating electrical machines. ElectricPower Applications, IET. 2008;2(4):215-247.

61. Baker PC, et al. Development of an integrated low-power RF partial dischargedetector. Electrical Insulation Conference 2009. EIC 2009, IEEE; 2009.

62. Tetrault, Serge M, Greg C Stone, Howard G, Sedding. Monitoring partial dischargeson 4-kV motor windings. Industry Applications, IEEE Transactions. 1999;35(3):682-688.

63. Stone, Greg, et al. Electrical insulation for rotating machines: design, evaluation,aging, testing, and repair. Vol. 21. Wiley-IEEE Press; 2004.

64. Toliyat, Hamid A, Gerald B. Kliman, eds. Handbook of electric motors. Vol. 120. CRCPress; 2010.

65. Bishop, Tom. Squirrel cage rotor testing. EASA Convention; 2003.66. Testing of Squirrel Cage Rotors, Elect. App. Serv. Assoc., Inc., St. Louis, MO, EASA

Tech. Note 23, 2003.67. McKinnon, David L. Using a six fault zone approach for predictive maintenance on

motors. Electrical Insulation Conference and Electrical Manufacturing Expo. IEEE;2007.

68. El Menzhi L, Saad A. Induction motor fault diagnosis using voltage Park componentsof an auxiliary winding-voltage unbalance. Electrical Machines and Systems, 2009.ICEMS 2009. International Conference. IEEE; 2009.

69. El Menzhi, Lamiaa, Abdallah Saad. Induction motor fault diagnosis using voltagespectrum of an auxiliary winding. Electrical Machines and Systems, 2007. ICEMS.International Conference on. IEEE; 2007.


332

70. Mirimani, Seyyed Mehdi, et al. Static Eccentricity Fault Detection in Single Stator-Single Rotor Axial Flux Permanent Magnet Machines. Industrial Application, IEEETransaction. 2011;48(6):1-8.

71. Acosta GG, Verucchi CJ, Gelso ER, A current monitoring system for diagnosingelectrical failures in induction motors, Mechanical Systems and Signal Processing.2006;20:953–965.

72. Urresty J, et al. Detection of demagnetization faults in surface-mounted permanentmagnet synchronous motors by means of the zero-sequence voltage component.Energy Conversion, IEEE Transactions. 2012;27(1):42-51.

73. Urresty J-C, et al. Detection of inter-turns short circuits in permanent magnetsynchronous motors operating under transient conditions by means of the zerosequence voltage. Power Electronics and Applications (EPE 2011), Proceedings ofthe 2011-14th European Conference. IEEE; 2011.

74. Briz, Fernando, et al. Induction machine diagnostics using zero sequence component.Industry Applications Conference, 2005. Fourtieth IAS Annual Meeting. ConferenceRecord of the 2005. Vol. 1. IEEE; 2005.

75. Scheffer, Cornelius, Paresh Girdhar. Practical machinery vibration analysis andpredictive maintenance. Newnes; 2004.

76. Han Y, Song YH. Condition monitoring techniques for electrical equipment-a literaturesurvey. Power Delivery, IEEE Transactions. 2003;18(1):4-13.

77. Rao JS. Vibratory condition monitoring of machines. CRC PressI Llc; 2000.78. Tavner, Peter J, James Penman. Condition monitoring of electrical machines.

Letchworth: Research Studies Press; 1987.79. Pöyhönen, S., et al. Numerical magnetic field analysis and signal processing for fault

diagnostics of electrical machines. COMPEL: The International Journal forComputation and Mathematics in Electrical and Electronic Engineering. 2003;22(4):969-981.

80. Choi, Seungdeog, et al. Performance-oriented electric motors diagnostics in modernenergy conversion systems. Industrial Electronics, IEEE Transactions. 2012;59(2):1266-1277.

81. Pineda-Sanchez M, et al. Diagnosis of induction motor faults in time-varying conditionsusing the polynomial-phase transform of the current. Industrial Electronics, IEEETransactions. 2011;58(4):1428-1439.

82. Yang Yu, Cheng Junsheng. A roller bearing fault diagnosis method based on EMDenergy entropy and ANN. Journal of sound and vibration. 2006;294(1):269-277.

83. Lin J, Zuo MJ, Fyfe KR. Mechanical fault detection based on the wavelet denoisingtechnique. Journal of Vibration and Acoustic. 2004;126(1):9–16.

84. Dorrell, David G, William T. Thomson, and Steven Roach. Analysis of airgap flux,current, and vibration signals as a function of the combination of static and dynamicairgap eccentricity in 3-phase induction motors. Industry Applications, IEEETransactions. 1997;33(1):24-34.

85. Zhuge J. Vibration Data Collector Signal Analysis, Part of VDC User’s Manual,CRYSTAL instrument; 2009.

86. Finley, William R, Mark M. Hodowanec, and Warren G. Holter. An analytical approachto solving motor vibration problems. Petroleum and Chemical Industry Conference,1999. Industry Applications Society 46th Annual. IEEE; 1999.

87. Cameron JR, Thomson WT, Dow AB. Vibration and current monitoring for detectingairgap eccentricity in large induction motors. IEE Proceedings B (Electric PowerApplications). Vol. 133. No. 3. IET Digital Library; 1986.


333

88. Rodríguez, Pedro Vicente Jover, et al. Air-gap force distribution and vibration patternof induction motors under dynamic eccentricity. Electrical Engineering.2008;90(3):209-218.

89. Ellison AJ, Yang SJ. Effects of rotor eccentricity on acoustic noise from inductionmachines. Proceedings of the Institution of Electrical Engineers. Vol. 118. No. 1. IETDigital Library; 1971.

90. Sameh M, Metwalley, Shawki A. Abouel-seoud. Condition based maintenanceoptimization for faulty gearbox under continuous noise monitoring. British Journal ofApplied Science & Technology. 2013;3(2):320-339.

91. Baydar, Naim, Andrew Ball. A comparative study of acoustic and vibration signals indetection of gear failures using Wigner–Ville distribution. Mechanical systems andsignal processing. 2001;15(6):1091-1107.

92. Scanlon, Patricia, Darren F, Kavanagh, Francis M. Boland. Residual Life Prediction ofRotating Machines Using Acoustic Noise Signals. 2013;1-14.

93. Scanlon, Patricia, Alan M, Lyons, Alan O'Loughlin. Acoustic signal processing fordegradation analysis of rotating machinery to determine the remaining useful life.Applications of Signal Processing to Audio and Acoustics, 2007 IEEE Workshop.IEEE; 2007.

94. Shahin Hedayati Kia, Humberto Henao, Capolino G-A. Mechanical transmission andtorsional vibration effects on induction machine stator current and torque in railwaytraction systems. Energy Conversion Congress and Exposition, 2009. ECCE 2009.IEEE. IEEE; 2009.

95. Holtz, Joachim. Sensorless control of induction motor drives. Proceedings of the IEEE.2002;90(8):1359-1394.

96. Henao, Humberto, Shahin Hedayati Kia, Capolino G. Torsional-vibration assessmentand gear-fault diagnosis in railway traction system. Industrial Electronics, IEEETransactions. 2011;58(5):1707-1717.

97. Guzinski, Jaroslaw, et al. Application of speed and load torque observers in high-speed train drive for diagnostic purposes. Industrial Electronics, IEEE Transactions.2009;56(1):248-256.

98. Mehala NEELAM. Condition monitoring and fault diagnosis of induction motor usingmotor current signature analysis. Doctor of philosophy, Electrical Engineering,National Institute of Technology Kurukshetra, India October 2010.

99. Hsu, John S. Monitoring of defects in induction motors through air-gap torqueobservation. Industry Applications, IEEE Transactions. 1995;31(5):1016-1021.

100. R. Maier. Protection of squirrel-cage induction motors using instantaneous power andphase application. Industry Application s, IEEE Transaction. 1992;28(2):376–380.

101. Motor and Generators, NEMA Standard Publication No.MG 1- 1998 (Revision 2-2001).102. Carson CC, Barton SC, Echeverria FS. Immediate Warning of Local Overheating in

Electric Machines by the Detection of Pyrolysis Products. Power Apparatus andSystems, IEEE Transactions. 1973;2:533-542.

103. Hongbo, Fan, et al. Study on oil detection technology based on inductive wear debrissensor. Electronic Measurement & Instruments, 2009. ICEMI'09. 9th InternationalConference. IEEE; 2009.

104. Xufeng, Jiang, et al. Study on Aero-engine Wear Fault Diagnosis by Direct ReadingFerrograph. Measuring Technology and Mechatronics Automation (ICMTMA), 2011.Third International Conference. Vol. 3. IEEE; 2011.


334

105. Laghari MS, Ahmed F, Aziz J. Wear particle shape and edge detail analysis.Computer and Automation Engineering (ICCAE), 2010 The 2nd InternationalConference. Vol. 1. IEEE; 2010.

106. Lokesh A, Gupta, Dimitrios Peroulis. Wireless temperature sensor for conditionmonitoring of bearings operating through thick metal plates. IEEE Sensor Journal.2013;13(6):2292-2298.

107. Milic, Saša D, Milesa Z. Sreckovic. A stationary system of noncontact temperaturemeasurement and hotbox detecting. Vehicular Technology, IEEE Transactions.2008;57(5):2684-2694.

108. Boglietti, Aldo, et al. Evolution and modern approaches for thermal analysis ofelectrical machines. Industrial Electronics, IEEE Transactions. 2009;56(3):871-882.

109. Mellor PH, Roberts D, Turner DR. Lumped parameter thermal model for electricalmachines of TEFC design. IEE Proceedings B (Electric Power Applications). Vol. 138.No. 5. IET Digital Library; 1991.

110. Okoro, Ogbonnaya I. Steady and transient states thermal analysis of a 7.5-kWsquirrel-cage induction machine at rated-load operation. Energy Conversion, IEEETransactions. 2005;20(4):730-736.

111. Nerg, Janne, Marko Rilla, Juha Pyrhonen. Thermal analysis of radial-flux electricalmachines with a high power density. Industrial Electronics, IEEE Transactions.2008;55(10):3543-3554.

112. Chowdhury SK, Baski PK. A simple lumped parameter thermal model for electricalmachine of TEFC design. Power Electronics, Drives and Energy Systems (PEDES) &2010 Power India, 2010 Joint International Conference. IEEE; 2010.

113. Gao Zhi, Thomas G. Habetler, Ronald G, Harley. An online adaptive stator windingtemperature estimator based on a hybrid thermal model for induction machines.Electric Machines and Drives, 2005 IEEE International Conference. IEEE; 2005.

114. Moreno J, Fernández F, Pérez Hidalgo, Martinez MD. Realisation of tests to determinethe parameters of the thermal model of an induction machine. Electric PowerApplications, IEE Proceedings. Vol. 148. No. 5. IET; 2001.

115. Milanfar, Peyman, Jeffrey H, Lang. Monitoring the thermal condition of permanent-magnet synchronous motors. Aerospace and Electronic Systems, IEEE Transactions.1996;32(4):1421-1429.

116. Jankowski, Todd A, et al. Development and validation of a thermal model for electricinduction motors. Industrial Electronics, IEEE Transactions. 2010;57(12):4043-4054.

117. Nategh, Shafigh, et al. Thermal analysis of a PMaSRM using partial FEA and lumpedparameter modeling. Energy Conversion, IEEE Transactions. 2012;27(2):477-488.

118. Kolondzovski Z, Belahcen A, Arkkio A. Comparative thermal analysis of different rotortypes for a high-speed permanent-magnet electrical machine. Electric PowerApplications, IET. 2009;3(4):279-288.

119. El-Refaie, Ayman M, et al. Thermal analysis of multibarrier interior PM synchronousmachine using lumped parameter model. Energy Conversion, IEEE Transactions.2004;19(2):303-309.

120. Idoughi, Laïd, et al. Thermal model with winding homogenization and FITdiscretization for stator slot. Magnetics, IEEE Transactions. 2011;47(12):4822-4826.

121. Lipo, Thomas A. Vector control and dynamics of AC drives. Vol. 41. Oxford UniversityPress, USA; 1996.

122. Sang-Bin Lee, et al. A stator and rotor resistance estimation technique for conductortemperature monitoring. Industry Applications Conference, 2000. Conference Recordof the 2000 IEEE. Vol. 1. IEEE; 2000.


335

123. Zhang, Pinjia, et al. A DC signal injection-based thermal protection scheme for soft-starter-connected induction motors. Industry Applications, IEEE Transactions.2009;45(4):1351-1358.

124. Zhang, Pinjia, Bin Lu, Thomas G, Habetler. A remote and sensorless stator windingresistance estimation method for thermal protection of soft-starter-connected inductionmachines. Industrial Electronics, IEEE Transactions. 2008;55(10):3611-3618.

125. Cheng, Siwei, et al. A nonintrusive thermal monitoring method for closed-loop drive-fed induction machines. Energy Conversion Congress and Exposition (ECCE), 2011IEEE. IEEE; 2011.

126. Samanta B, Nataraj C. Use of particle swarm optimization for machinery faultdetection. Engineering Applications of Artificial Intelligence. 2009;22(2):308-316.

127. Nejjari, Hamid, Mohamed El Hachemi Benbouzid. Monitoring and diagnosis ofinduction motors electrical faults using a current Park's vector pattern learningapproach. Industry Applications, IEEE Transactions. 2000:36(3):730-735.

128. Wang, Huaqing, Peng Chen. Sequential fuzzy diagnosis for condition monitoring ofrolling bearing based on neural network. Advances in Neural Networks-ISNN 2008.Springer Berlin Heidelberg. 2008;284-293.

129. Tag Eldin EM, et al. Monitoring and diagnosis of external faults in three phaseinduction motors using artificial neural network. Power Engineering Society GeneralMeeting, 2007. IEEE. IEEE; 2007.

130. Asiri, Yahya, et al. Neural network based classification of partial discharge in HVmotors. Electrical Insulation Conference (EIC), 2011. IEEE, 2011.

131. Raj, Santhana, Murali N. Early classification of bearing faults using morphologicaloperators and fuzzy inference. Industrial Electronics, IEEE Transaction. 2013;60(2):567-574.

132. Chandra Sekhar S, Sreenivas TV. Adaptive spectrogram vs. adaptive pseudo-Wigner–Ville distribution for instantaneous frequency estimation. Signal Processing.2003;83(7):1529-1543.

133. Cayrac, Didier, Didier Dubois, Henri Prade. Handling uncertainty with possibility theoryand fuzzy sets in a satellite fault diagnosis application. Fuzzy Systems, IEEETransactions. 1996;4(3):251-269.

134. Li Ke, Peng Chen, Shiming Wang. An Intelligent diagnosis method for rotatingmachinery using least squares mapping and a fuzzy neural network. Sensors.2012;12(5):5919-5939.

135. Yazidi, Amine, et al. A web-based remote laboratory for monitoring and diagnosis ofac electrical machines. Industrial Electronics, IEEE Transactions. 2011;58(10):4950-4959.

136. Yazici, Birsen, Gerald B, Kliman. An adaptive statistical time-frequency method fordetection of broken bars and bearing faults in motors using stator current. IndustryApplications, IEEE Transactions. 1999;35(2):442-452.

137. Ocak, Hasan. Fault detection, diagnosis and prognosis of rolling element bearings:Frequency domain methods and hidden markov modeling. Diss. Case WesternReserve University; 2004.

138. Blodt M, Bonacci D, Regnier J, Chabret M, Faucher J. On-line monitoring ofmechanical faults in variable-speed induction motor drives using the Wingerdistribution. Industrial Electronics. IEEE Transaction. 2008;55(2):522-533.


336

APPENDIX

Time-Frequency Analysis method equations which discussed at this paper are explained inthis session.

1) Short-Time Fourier Transform (STFT):

The short-time Fourier transform (STFT) [41] by breaking signal into short blocks andapplying an FFT to each part can determine the sinusoidal frequency and phase componentof the its local time domain.

Mathematically, the STFT of a signal )(tx is explained as follows [42]:

dtjthtxtSTFTx )exp()()(21),(

(3)

In the above equation ω is an angular frequency, and )(h is the window function. With thetechnique of windowing (such as Gaussian, Hamming, Hanning …), the STFT can provideinformation about both time and frequency of the signal, since the time-varying concentrationinformation is required for real-time applications. STFT analysis may lose the transient andtemporal information and it is not good, but the STFT is simpler than the other methods. TheSTFT spectrum can be defined as follows [40]:

djthxtSTFTtP xx )exp()()(21),(),( 2

(4)

Of course other studies [137,138] showed that the techniques such as short-time Fouriertransform, where a nonstationary signal is divided into short pseudo-stationary segments,are not suitable for the analysis of signals with complex time–frequency characteristics.

2) Wavelet Analysis (WA)

WA is another time-frequency signal analysis method that has been widely used anddeveloped recent decade. It has the local characteristic of the time domain as well as thefrequency domain, and its time-frequency window is changeable. The Continuous WaveletTransform (CWT) of )(tx is a timescale method of signal processing that can be definedmathematically as the sum over all time of the signal multiplied by scaled and shiftedversions of the wavelet function )(t [42]:

dtabttx

abaCWTx

)()(1),( * Rba , (5)

Where )(* t is the complex conjugate of which denotes the mother wavelet or basicwavelet. a & b are parameters related to scale and time respectively. If a is small, higher-


337

frequency components can be analyzed, and when it is large, lower-frequency componentscan be analyzed. When b is given a value, the fundamental function can be shifted by adistance in the direction in which time advances. The CWT spectrum is considered asfollows. Wavelet transform has the isometric characteristic.

3) Winger-Ville Distribution (WVD):

The Wigner-Ville Distribution (WVD) [41] is a very important quadratic-form time-frequencydistribution with optimized resolution in both the time and frequency domains. The WVD ismatched to linear chirps and can represent it effectively. The instantaneous frequency ofsuch signals can be estimated easily by picking the peak in the time-frequency plane 40.However, the WVD does not yield a localized distribution for frequency variations that arenot linear [44,133].

The instantaneous frequency within the window can be considered to be nearly linearbecause the VWD variants need windowing.

The Pseudo-Wigner-Ville distribution (PWVD) has better resolution and provides a moreaccurate estimate of the instantaneous frequency. Therefore, it has been used extensively invarious applications to display time-frequency spectral information [17]. The PWVD equationdefined as follows [98,132]:

dehtxtxtPWVD jx )()

21()

21(

21),( *

(6)

In this equation is an angular frequency and )(h is the windows function.

detststW j)21()

21(

21),( *

desS js)

21()

21(

21 *

(7)

Winger-Ville distribution of a motor in healthy condition and with faulty bearing is shown atFig. 10 [98].

(a)


338

(b)

Fig. 10. Winger-Ville distribution of motor (a) in healthy condition (b) Winger-Villedistribution of motor with faulty bearing [98]

_________________________________________________________________________© 2014 Mortazavizadeh and Mousavi; This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,distribution, and reproduction in any medium, provided the original work is properly cited.

Peer-review history:The peer review history for this paper can be accessed here:

http://www.sciencedomain.org/review-history.php?iid=327&id=33&aid=2538

a review on condition monitoring and diagnostic techniques of rotating electrical machines

Documents

internal faults

thesources of motor

commonmachine faults

secondary faults

combined methods

fault detection methods

stator current

faults5 stator phase